Merge branch 'sycl' into przemek/fix-vk-interop-on-igpu

Author: przemektmalon

.github/workflows/sycl-linux-build.yml

@@ -32,6 +32,10 @@ on:
       build_artifact_suffix:
         type: string
         required: true
+      build_target:
+        type: string
+        required: false
+        default: sycl-toolchain
       artifact_archive_name:
         type: string
         default: llvm_sycl.tar.zst
@@ -177,7 +181,8 @@ jobs:
           -DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=SPIRV
     - name: Compile
       id: build
-      run: cmake --build $GITHUB_WORKSPACE/build --target sycl-toolchain
+      # Emulate default value for manual dispatch as we've run out of available arguments.
+      run: cmake --build $GITHUB_WORKSPACE/build --target ${{ inputs.build_target || 'sycl-toolchain' }}
     - name: check-llvm
       if: always() && !cancelled() && contains(inputs.changes, 'llvm')
       run: |

.github/workflows/sycl-nightly.yml

@@ -33,6 +33,7 @@ jobs:
       build_cache_suffix: sprod_shared
       build_artifact_suffix: sprod_shared
       build_configure_extra_args: '--shared-libs --hip --cuda --native_cpu --no-assertions'
+      build_target: all
 
       artifact_archive_name: sycl_linux_shared.tar.zst
 
@@ -151,6 +152,7 @@ jobs:
       artifact_archive_name: sycl_windows.tar.gz
       # Disable the spirv-dis requirement as to not require SPIR-V Tools.
       build_configure_extra_args: -DLLVM_SPIRV_ENABLE_LIBSPIRV_DIS=off
+      build_target: all
 
   e2e-win:
     needs: build-win

.github/workflows/sycl-windows-build.yml

@@ -10,6 +10,10 @@ on:
       build_configure_extra_args:
         type: string
         required: false
+      build_target:
+        type: string
+        required: false
+        default: sycl-toolchain
       changes:
         type: string
         description: 'Filter matches for the changed files in the PR'
@@ -58,6 +62,11 @@ on:
       build_configure_extra_args:
         type: string
         required: false
+      build_target:
+        type: choice
+        options:
+          - "sycl-toolchain"
+          - "all"
       artifact_archive_name:
         type: choice
         options:
@@ -143,7 +152,7 @@ jobs:
       id: build
       shell: bash
       run: |
-        cmake --build build --target sycl-toolchain
+        cmake --build build --target ${{ inputs.build_target }}
     - name: check-llvm
       if: always() && !cancelled() && contains(inputs.changes, 'llvm')
       shell: bash

.github/workflows/ur-build-hw.yml

@@ -164,7 +164,8 @@ jobs:
     - name: Test adapter specific
       env:
         ZE_ENABLE_LOADER_DEBUG_TRACE: 1
-      run: ctest -C ${{matrix.build_type}} --test-dir build --output-on-failure -L "adapter-specific" -E "memcheck" --timeout 600 -VV
+        LIT_OPTS: "--timeout 120 --filter-out 'adapters/level_zero/memcheck.test'"
+      run: cmake --build build -j $(nproc) -- check-unified-runtime-adapter
       # Don't run adapter specific tests when building multiple adapters
       if: ${{ matrix.adapter.other_name == '' }}
 

clang/lib/Driver/ToolChains/Cuda.cpp

@@ -966,6 +966,8 @@ void CudaToolChain::addClangTargetOptions(
     if (FastRelaxedMath || UnsafeMathOpt)
       CC1Args.append({"-mllvm", "--nvptx-prec-divf32=0", "-mllvm",
                       "--nvptx-prec-sqrtf32=0"});
+
+    CC1Args.append({"-mllvm", "-enable-memcpyopt-without-libcalls"});
   } else {
     CC1Args.append({"-fcuda-is-device", "-mllvm",
                     "-enable-memcpyopt-without-libcalls",

clang/test/Driver/sycl-nvptx-memcpy-opt.cpp

@@ -0,0 +1,5 @@
+// RUN: %clang -### -nocudalib \
+// RUN:   -fsycl -fsycl-targets=nvptx64-nvidia-cuda %s 2>&1 \
+// RUN: | FileCheck --check-prefix=CHECK-DEFAULT %s
+
+// CHECK-DEFAULT: "-enable-memcpyopt-without-libcalls"

devops/dependencies.json

@@ -1,15 +1,15 @@
 {
   "linux": {
     "compute_runtime": {
-      "github_tag": "25.13.33276.19",
-      "version": "25.13.33276.19",
-      "url": "https://github.com/intel/compute-runtime/releases/tag/25.13.33276.19",
+      "github_tag": "25.18.33578.6",
+      "version": "25.18.33578.6",
+      "url": "https://github.com/intel/compute-runtime/releases/tag/25.18.33578.6",
       "root": "{DEPS_ROOT}/opencl/runtime/linux/oclgpu"
     },
     "igc": {
-      "github_tag": "v2.10.10",
-      "version": "v2.10.10",
-      "url": "https://github.com/intel/intel-graphics-compiler/releases/tag/v2.10.10",
+      "github_tag": "v2.11.7",
+      "version": "v2.11.7",
+      "url": "https://github.com/intel/intel-graphics-compiler/releases/tag/v2.11.7",
       "root": "{DEPS_ROOT}/opencl/runtime/linux/oclgpu"
     },
     "cm": {

libdevice/cmake/modules/SYCLLibdevice.cmake

@@ -1,3 +1,4 @@
+include(CheckCXXCompilerFlag)
 set(obj_binary_dir "${CMAKE_LIBRARY_OUTPUT_DIRECTORY}")
 set(obj-new-offload_binary_dir "${CMAKE_LIBRARY_OUTPUT_DIRECTORY}")
 if (MSVC)
@@ -344,6 +345,7 @@ if("native_cpu" IN_LIST SYCL_ENABLE_BACKENDS)
           COMPONENT libsycldevice)
 endif()
 
+check_cxx_compiler_flag(-Wno-invalid-noreturn HAS_NO_INVALID_NORETURN_WARN_FLAG)
 # Add all device libraries for each filetype except for the Intel math function
 # ones.
 add_devicelibs(libsycl-itt-stubs
@@ -358,7 +360,9 @@ add_devicelibs(libsycl-itt-user-wrappers
 
 add_devicelibs(libsycl-crt
   SRC crt_wrapper.cpp
-  DEPENDENCIES ${crt_obj_deps})
+  DEPENDENCIES ${crt_obj_deps}
+  EXTRA_OPTS $<$<BOOL:${HAS_NO_INVALID_NORETURN_WARN_FLAG}>:-Wno-invalid-noreturn>)
+
 add_devicelibs(libsycl-complex
   SRC complex_wrapper.cpp
   DEPENDENCIES ${complex_obj_deps})

libdevice/crt_wrapper.cpp

@@ -51,13 +51,24 @@ int memcmp(const void *s1, const void *s2, size_t n) {
 DEVICE_EXTERN_C_INLINE
 int rand() {
   size_t gid =
+#if defined(__NVPTX__) || defined(__AMDGCN__)
+      (__spirv_GlobalInvocationId_x() * __spirv_GlobalSize_y() *
+       __spirv_GlobalSize_z()) +
+      (__spirv_GlobalInvocationId_y() * __spirv_GlobalSize_z()) +
+      __spirv_GlobalInvocationId_z();
+#else
       (__spirv_BuiltInGlobalInvocationId.x * __spirv_BuiltInGlobalSize.y *
        __spirv_BuiltInGlobalSize.z) +
       (__spirv_BuiltInGlobalInvocationId.y * __spirv_BuiltInGlobalSize.z) +
       __spirv_BuiltInGlobalInvocationId.z;
-  size_t global_size = __spirv_BuiltInGlobalSize.x *
-                       __spirv_BuiltInGlobalSize.y *
-                       __spirv_BuiltInGlobalSize.z;
+#endif
+  size_t global_size =
+#if defined(__NVPTX__) || defined(__AMDGCN__)
+      __spirv_GlobalSize_x() * __spirv_GlobalSize_y() * __spirv_GlobalSize_z();
+#else
+      __spirv_BuiltInGlobalSize.x * __spirv_BuiltInGlobalSize.y *
+      __spirv_BuiltInGlobalSize.z;
+#endif
   size_t gid1 =
       (global_size > RAND_NEXT_LEN) ? (gid & (RAND_NEXT_LEN - 1)) : gid;
   if (RAND_NEXT_ACC[gid1] == 0)
@@ -73,13 +84,24 @@ int rand() {
 DEVICE_EXTERN_C_INLINE
 void srand(unsigned int seed) {
   size_t gid =
+#if defined(__NVPTX__) || defined(__AMDGCN__)
+      (__spirv_GlobalInvocationId_x() * __spirv_GlobalSize_y() *
+       __spirv_GlobalSize_z()) +
+      (__spirv_GlobalInvocationId_y() * __spirv_GlobalSize_z()) +
+      __spirv_GlobalInvocationId_z();
+#else
       (__spirv_BuiltInGlobalInvocationId.x * __spirv_BuiltInGlobalSize.y *
        __spirv_BuiltInGlobalSize.z) +
       (__spirv_BuiltInGlobalInvocationId.y * __spirv_BuiltInGlobalSize.z) +
       __spirv_BuiltInGlobalInvocationId.z;
-  size_t global_size = __spirv_BuiltInGlobalSize.x *
-                       __spirv_BuiltInGlobalSize.y *
-                       __spirv_BuiltInGlobalSize.z;
+#endif
+  size_t global_size =
+#if defined(__NVPTX__) || defined(__AMDGCN__)
+      __spirv_GlobalSize_x() * __spirv_GlobalSize_y() * __spirv_GlobalSize_z();
+#else
+      __spirv_BuiltInGlobalSize.x * __spirv_BuiltInGlobalSize.y *
+      __spirv_BuiltInGlobalSize.z;
+#endif
   size_t gid1 =
       (global_size > RAND_NEXT_LEN) ? (gid & (RAND_NEXT_LEN - 1)) : gid;
   RAND_NEXT_ACC[gid1] = seed;

libdevice/spirv_vars.h

@@ -10,11 +10,23 @@
 #define __LIBDEVICE_SPIRV_VARS_H
 
 #include "device.h"
+#include <cstddef>
 
-#if defined(__SPIR__) || defined(__SPIRV__) || defined(__NVPTX__) ||           \
-    defined(__AMDGCN__)
+#if defined(__NVPTX__) || defined(__AMDGCN__)
+// For AMD/Cuda those symbols will be provided by libclc.
+DEVICE_EXTERNAL size_t __spirv_GlobalInvocationId_x();
+DEVICE_EXTERNAL size_t __spirv_GlobalInvocationId_y();
+DEVICE_EXTERNAL size_t __spirv_GlobalInvocationId_z();
+DEVICE_EXTERNAL size_t __spirv_LocalInvocationId_x();
+DEVICE_EXTERNAL size_t __spirv_LocalInvocationId_y();
+DEVICE_EXTERNAL size_t __spirv_LocalInvocationId_z();
+DEVICE_EXTERNAL size_t __spirv_GlobalSize_x();
+DEVICE_EXTERNAL size_t __spirv_GlobalSize_y();
+DEVICE_EXTERNAL size_t __spirv_GlobalSize_z();
+#endif // __NVPTX__ || __AMDGCN__
+
+#if defined(__SPIR__) || defined(__SPIRV__)
 
-#include <cstddef>
 #include <cstdint>
 
 #define __SPIRV_VAR_QUALIFIERS EXTERN_C const
@@ -63,5 +75,5 @@ const size_t_vec __spirv_BuiltInGlobalInvocationId{};
 const size_t_vec __spirv_BuiltInLocalInvocationId{};
 #endif // !__SPIR__ && !__SPIRV__
 
-#endif // __SPIR__ || __SPIRV__ || __NVPTX__ || __AMDGCN__
+#endif // __SPIR__ || __SPIRV__
 #endif // __LIBDEVICE_SPIRV_VARS_H

llvm/lib/SYCLPostLink/CMakeLists.txt

@@ -22,8 +22,10 @@ add_llvm_component_library(LLVMSYCLPostLink
   BitWriter
   Core
   Demangle
+  InstCombine
   IRPrinter
   Passes
+  ScalarOpts
   Support
   ipo
   )